Talk Morgan Forums Community The Malvern Arms Artificial Intelligence

Hmmmmm.
How far is effective autonomous driving away? I mean for everyone not in restricted areas. Tesla are mining data from their cars to build up enough data to make it work. A lot of scenarios for the system to monitor, analyse, solve. Any Tesla owners here? Are they doing auto software updates? If so how do you get told?
I have a Kia with “smart” lane assist, reactive braking.
In the manual there are pages of warnings about inaccuracies in the lane assist. False readings from verges, roadside furniture, bends, faded road markings etc. I find it giving false warnings on many roads, bends, worn surfaces.

My Mercedes comes standard with a camera fitted in the front screen and sat nav, it flashes every speed limit change even temporary speed limit changes to my central instrument cluster, it is unerringly accurate and has to-date recognised every change in speed limit. It would be quite easy to, collect that data along with my actual speed for enforcement agencies to act upon it, should the requisite legislation allow it..

I had a Tesla Model 3 Dual motor from early 2021 until mid 2022 and drove 11,000 miles.
The autonomous driving was, frankly, useless and occasionally dangerous, phantom braking when it thought it saw something, over reaction to potential hazards, and frequent mis reading of speed limit signs, mostly due to missing or obscured signs. Likewise, faded or missing white lines caused it to stop working. in total it made driving stressful and not enjoyable.
Downloads came without warning, most didn't appear to add anything useful.

In the end I disabled it, the software was one of the reasons I got rid of the car.

My view? Autonomous is a long way from being viable.

Machine Learning/Artificial Intelligence flies in the face of good science and is therefore open to abuse.
Back in the early 1970s when I was learning how to be an inquisitive scientists we had a compulsory unit, "Fundamentals of Science".
The message was simple: believe nothing that you cannot verify for yourself.
It is a rule that I've applied to not just research science, but to much of life however it does make on a bit cynical....

To me things like autonomous lane keeping/braking etc are just clever progaming and access to high quality sensors rather than AI

When my robot lawnmower disappears and I get an email from a beach in Goa saying it was pissed off with its life and needs to find itself ... then I'll acknowledge AI ... AI should have genuine internal agency ... the term is being bandied about (much the same as a GT badge in the old-time car world) as a marketing ploy ... 'real AI' will be a lot less useful than we think .... because it also thinks rather than just does what it's told, so is as fundamentally untrustworthy as most Humans!

K

There are interesting variations of the definition of AI, but the one I like is the theory and development of computer systems able to perform tasks normally requiring human intelligence, such as visual perception, speech recognition, decision-making, and translation between languages. Most of my Mercedes command system seems to meet that criteria, with the exception of translation between languages, having said that it can operate in various languages once selected. It certainly intervenes (decision-making) in a given set of circumstances to take control of the car eg detects the driver asleep at the wheel. Whilst it can recognise my individual driving style, and adjusts/anticipates my style of driving from acceleration response, gear changing I'm not sure if it learns and adapts its prime programming beyond it's original programming parameters and operating envelope. Its that element of AI that is the most of concern to me that is the ability to write its own software and therefore determine its own operating envelope..

AI counter suing AI, gets stuck in a vicious loop and the internet stops

One thing AI will learn quickly is how to mimic human lawyers and charge like wounded rhinos